🚀 go-pugleaf

Hi Friends

Here is the summary from all the responses that I got regarding  
Number-Crunching on the NeXT:

The SPARC 2 is slightly faster than the NeXT Turbo when it comes to scientific  
engineering applications. With an application compiled with one of the Fortran  
Compilers, the NeXT is as good as a SPARC 2. Since the current C compiler and  
libraries are not very efficient, when it comes to transcedental functions(Sin,  
Cos etc), applications written in C run much slower. 

The Conclusion: We will be going for a NeXT Turbo.

Here is what the other users had to say:

*******************************************************
From: "Christopher Penrose" <penrose@silvertone.Princeton.EDU>

 I have used Sparcstation 1+ 's and a 25MHz 68040 NeXT is
comparable in  performance.  I have used intensive signal
processing algorithms (FFTs)  repeatedly.  Given the
same number crunching performance, I would definitely 
choose the NeXT.  Its operating system advantages (Mach)
and its window and development environment (NeXTstep)
have many advantages over Solaris and their typical
kludgey X-windows environments. 
 

Christopher

**************************************************************
From: m@crito.Stanford.EDU (M Carling)

A 33MHz NeXT is in the same class as a SPARC II. The Sun might be 10% faster.

M Carling
Director, Bay Area NeXT Group

**************************************************************
From: <Steve.Johnson@math.tamu.edu>

I have used both and have found the Sun to be up to 50% faster.  The
NeXT does very poorly on transcendentals (sin, cos, ...), so Fourier
transforms and other apps are quite slow.  If you're really looking
for speed, look at the HP & IBM boxes.  However, I'm not at all impressed
by their respective OS's.

Standard disclaimers, as usual.

- Steve Johnson, Systems Manager, Dept of Mathematics, Texas A&M University

**************************************************************
From: k0s8160@chenext1.tamu.edu (Kenji Seto)

I can't compare it with a Sun, but I am running an extremely large simulation,
on the NeXT and a Cray-YMP.  For CPU equivalence,
       1 Cray cpu-hour   =  20 Next 040-turbo hours

This is only approximate, +/- 3 hours.
Hope this helps....Kenji

*************************************************************
From: matt@zeb.ame.nd.edu (Matthew J. Grismer)

Using the Absoft Fortran compiler, I have achieved about the same performance  
as a Sparc 2 with a NeXTstation (not turbo). This is for computational fluid  
dynamics types of applications. I do have one code that for some reason runs  
about as fast as a Sparc 1, but I didn't write it, so I am not sure what the  
difference is. If you use C, the current version of the C compiler that comes  
with the Next does not do trancendentals well, but I believe that is fixed in  
NeXTSTEP 3.0. Actually, I have heard that Sunworld magazine reviewed a  
NeXTstation turbo, and rated it higher overall than the Sparc 2. Overwall, I am  
quite pleased with the number crunching abilitity of my station. I hope this  
helps.

MJG
NeXTmail preferred

**************************************************************
From: tennant@alph.msfc.nasa.gov (Allyn Tennant)

The floating point specmarks show that the SPARC 2 machines are faster than
the NeXT, however, the NeXT is very competitive with the cheaper SPARC 1  
machines.
I believe the NeXTstation is slightly faster than the IPC and of course, the
turbo is slightly faster yet.  This agrees with my experience with running
my software on the two machines.

Of course, if you want speed you should get an HP computer. 

For Mathematica the NeXT is VERY fast.  We find that Mathematica on a  
NeXTstation
is 10-20% faster than an IBM RS6000 model 320.  We do not have Mathematica on a
SUN so we can't compare, however, I've seen a couple of reports on the Net that
indicate that the NeXT is faster.

My work mostly involves time series analysis of data collect from X-ray
astronomy satellites.  I use the Oasys Green Hill Fortran compiler which is
fairly good and supports most VMS extensions.

					Allyn

**************************************************************
From: Nicolas Strobel <strobel@dirac.phys.washington.edu>

Depends on what kind of floating point operations you're wanting to do.
Transcendentals, and some other floating point operations are trapped in
the kernel and are emulated through software (so they're slow), but you
can get around that by using lookup tables.  These are the floating point
instructions done in *hardware* on the 040:
 
fmove, fsmove,fcmp,fsabs,fsnet,fadd,fdiv,fbcc,fdbcc,fsadd,fsmul,fdadd,fdmul,
fsqrt,fdsqrt,fsave,fmovem,fdmove,fabs,fdabs,fneg,fdneg,fsub,fmul,fscc,ftrapcc,
fssub,fsdiv,fdsub,fddiv,fssqrt,frestore

I don't know what all of these do--I just have this list of things in front of
me--but you can see things like abs,add,subtract,multiply,divide,sqrt,negate.
The other floating point operations are emulated through an obsolete software
library (motorola has a newer better version that did not make it into
2.X).

For the hardware instructions the NeXT 040 (25 MHz) is faster than a Sparc 1+. 
The turbo compares well with a sparc 2.  Using lookup tables, people have 
speeded up the transcendentals (sin,cos,exp, etc.) evaluation tremendously.  
Some person posted a note to the net about a library of lookup tables for the 
transcendentals he created. I can check around for the posting, if you'd like
more info.

If you want to do some serious number crunching on a workstation, you should
look at IBM RS6000's, HP's with the PA-Risc chips, or Dec 5000's. If you
want a nice general purpose workstation that is friendly, intuitive 
plug-and-play, has award winning software, and alright for some number 
crunching, the NeXT is the one to use.

Nick
strobel@dirac.phys.washington.edu
***************************************************************

Ramesh

--